transformers
Add a default decoder_attention_mask for EncoderDecoderModel during training
#26752
Merged
Merged